Machine approach eliminates labor-intensive data entry

Tired of plugging patient data into computers for quality assessments, especially when it requires digging through files and other time-consuming tasks? Relief may be in sight.

Jamie E. Anderson, MD, MPH, of the surgery department at the University of California, San Diego, and colleagues published a study in the January issue of JAMA Surgery that tested a man vs. machine approach for collecting risk-adjustment data elements. Machine, in this case automated collection from EHRs of objective data, did just fine.

The researchers used data from the American College of Surgeons National Surgical Quality Improvement Program (NSQIP) database from 2005 to 2010. The database includes 135 variables on each patient enrolled with a 30-day follow-up. Their analysis looked at almost 750,000 patients treated by general, cardiothoracic, vascular and plastic surgeons.

They designed two models: one with 66 preoperative risk variables in the NSQIP and a second that relied on 25 variables that were laboratory data or could be extracted from the patient’s medical record. They calculated area under the receiver operating characteristic curve (AUC) based on objective and all variables to compare their performances.

The mortality rate and the rate of at least one complication totaled 2.8 percent and 15.8 percent, respectively. Colectomy and aortic valve replacement had the highest mortality rates, at 4.4 percent and 4.3 percent. Esophagectomy topped the list for complications (45.8 percent) followed by aortic valve replacement (35 percent).

The AUC was 0.9005 for all variables vs. 0.8774 for objective variables for 30-day mortality and 0.9078 vs. 0.8881 for inpatient mortality. For complications, AUC equaled 0.7401 for all variables vs. 0.7137 for objective variables; for 30-day complications, it was 0.7859 vs. 0.7609.

The difference in AUC was greatest for patients who underwent aortic valve replacement.

“The difference in AUC was lower when examining complications than mortality, although the range of difference in AUC values differed by procedure,” they wrote. “Thus, this method may be best to use when wanting to perform risk-adjusted analyses for complications.”

Removing the human element from the process likely would improve efficiencies, lower costs and eliminate any possible gaming of the system, Anderson et al added. “This approach also addresses common concerns about the burden of data collection and the validity and reliability of data elements and can lead to wider adoption of quality assessment systems in health care.”